Energy Saving of HVAC System Using Integrated Machine Learning Algorithm (IMLA)

¹ Cheongju University, CJU Industry-Academic Cooperation Foundation, 28503 Cheongju-si, Chungcheongbuk-do, Korea
² Daejin University, Department of Architectural Engineering, 11159 Pocheon-si, Gyeonggi-do, Korea
³ Kangwon National University, Department of Urban Architecture, 25913 Samcheok-si, Gangwon-do, Korea
⁴ Cheongju University, Department of Architectural Engineering, 28503 Cheongju-si, Chungcheongbuk-do, Korea

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it. ,kr

Abstract

Abstract. Heating, Ventilation, and Air Conditioning (HVAC) systems account for a substantial portion of building energy consumption, making their efficient operation a critical issue for energy conservation and carbon reduction. Conventional HVAC control strategies are generally based on static rules or predefined schedules, which often fail to respond effectively to dynamically varying thermal loads, occupant behavior, and outdoor conditions. To overcome these limitations, this study proposes an Integrated Machine Learning Algorithm (IMLA) that unifies short-term load prediction and global optimization within a single HVAC control framework. The proposed IMLA combines an Artificial Neural Network (ANN) and a Genetic Algorithm (GA). The ANN is employed to predict cooling and heating loads at the next time step (t+1), providing anticipatory information on future thermal demand, while the GA determines optimal HVAC control variables by minimizing total energy consumption. By integrating predictive and optimization modules, the proposed framework enables proactive and coordinated control of air-side and water-side HVAC systems. The IMLA was applied to a Medium Office Reference Building, where the HVAC system was modeled as a variable air volume (VAV) air-side system coupled with a chilled-water-based plant. Its performance during the cooling period was evaluated and compared with conventional rule-based control and GA-based optimization without load prediction. Simulation results demonstrated that the proposed IMLA consistently outperformed the benchmark strategies. Compared with conventional control, the IMLA reduced fan energy consumption by approximately 13.4%, chiller energy by 8.0%, and pump energy by 7.0%. When total HVAC energy consumption was considered, the proposed approach achieved an overall energy reduction of approximately 8.2%, exceeding the performance of optimization-only control. These results indicate that incorporating short-term load prediction into the optimization process provides additional system-level energy savings by enabling more proactive and demand-responsive HVAC operation. The proposed IMLA offers a practical and extensible solution for improving HVAC energy efficiency under dynamic operating conditions and shows strong potential for application to various HVAC system configurations and seasonal operating modes.